Hugging Face's logo

qualcomm
/
NAFNet-DeBlur

Image-to-Image
PyTorch
android
Model card Files
xet
 Community
qaihm-bot commited on
Commit
83cdeb0
·
verified ·
1 Parent(s): 21380d1

v0.55.0

Browse files

See https://github.com/qualcomm/ai-hub-models/releases/v0.55.0 for changelog.

Files changed (3) hide show
  1. LICENSE +1 -0
  2. README.md +140 -0
  3. release_assets.json +46 -0
LICENSE ADDED
@@ -0,0 +1 @@
 
 
1
+ The license of the original trained model can be found at https://github.com/megvii-research/NAFNet/blob/main/LICENSE.
README.md ADDED
@@ -0,0 +1,140 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: pytorch
3
+ license: other
4
+ tags:
5
+ - android
6
+ pipeline_tag: image-to-image
7
+
8
+ ---
9
+
10
+ ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_deblur/web-assets/model_demo.png)
11
+
12
+ # NAFNet-DeBlur: Optimized for Qualcomm Devices
13
+
14
+ NAFNET is designed for lightweight real-time deblurring of images.
15
+
16
+ This is based on the implementation of NAFNet-DeBlur found [here](https://github.com/megvii-research/NAFNet.git).
17
+ This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/nafnet_deblur) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).
18
+
19
+ Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device.
20
+
21
+ ## Getting Started
22
+ There are two ways to deploy this model on your device:
23
+
24
+ ### Option 1: Download Pre-Exported Models
25
+
26
+ Below are pre-exported model assets ready for deployment.
27
+
28
+ | Runtime | Precision | Chipset | SDK Versions | Download |
29
+ |---|---|---|---|---|
30
+ | ONNX | float | Universal | QAIRT 2.42, ONNX Runtime 1.25.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_deblur/releases/v0.55.0/nafnet_deblur-onnx-float.zip)
31
+ | ONNX | w8a16 | Universal | QAIRT 2.42, ONNX Runtime 1.25.0 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_deblur/releases/v0.55.0/nafnet_deblur-onnx-w8a16.zip)
32
+ | QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_deblur/releases/v0.55.0/nafnet_deblur-qnn_dlc-float.zip)
33
+ | QNN_DLC | w8a16 | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_deblur/releases/v0.55.0/nafnet_deblur-qnn_dlc-w8a16.zip)
34
+ | TFLITE | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_deblur/releases/v0.55.0/nafnet_deblur-tflite-float.zip)
35
+
36
+ For more device-specific assets and performance metrics, visit **[NAFNet-DeBlur on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/nafnet_deblur)**.
37
+
38
+
39
+ ### Option 2: Export with Custom Configurations
40
+
41
+ Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/nafnet_deblur) Python library to compile and export the model with your own:
42
+ - Custom weights (e.g., fine-tuned checkpoints)
43
+ - Custom input shapes
44
+ - Target device and runtime configurations
45
+
46
+ This option is ideal if you need to customize the model beyond the default configuration provided here.
47
+
48
+ See our repository for [NAFNet-DeBlur on GitHub](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/nafnet_deblur) for usage instructions.
49
+
50
+ ## Model Details
51
+
52
+ **Model Type:** Model_use_case.image_editing
53
+
54
+ **Model Stats:**
55
+ - Model checkpoint: NAFNet-REDS-width64
56
+ - Input resolution: 360x640
57
+ - Number of parameters: 67.89M
58
+ - Model size (float): 271.55 MB
59
+
60
+ ## Performance Summary
61
+ | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
62
+ |---|---|---|---|---|---|---
63
+ | NAFNet-DeBlur | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 46.752 ms | 8 - 991 MB | NPU
64
+ | NAFNet-DeBlur | ONNX | float | Snapdragon® X2 Elite | 50.135 ms | 162 - 162 MB | NPU
65
+ | NAFNet-DeBlur | ONNX | float | Snapdragon® X Elite | 114.344 ms | 146 - 146 MB | NPU
66
+ | NAFNet-DeBlur | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 82.212 ms | 9 - 1261 MB | NPU
67
+ | NAFNet-DeBlur | ONNX | float | Qualcomm® QCS8550 (Proxy) | 111.326 ms | 0 - 171 MB | NPU
68
+ | NAFNet-DeBlur | ONNX | float | Snapdragon® 8 Elite For Galaxy Mobile | 63.027 ms | 8 - 928 MB | NPU
69
+ | NAFNet-DeBlur | ONNX | float | Qualcomm® QCS9075 | 153.169 ms | 8 - 53 MB | NPU
70
+ | NAFNet-DeBlur | ONNX | float | Qualcomm® QCS8750 | 63.027 ms | 8 - 928 MB | NPU
71
+ | NAFNet-DeBlur | ONNX | float | Qualcomm® QCS7181 | 114.344 ms | 146 - 146 MB | NPU
72
+ | NAFNet-DeBlur | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 49.896 ms | 4 - 1202 MB | NPU
73
+ | NAFNet-DeBlur | ONNX | w8a16 | Snapdragon® X2 Elite | 52.661 ms | 211 - 211 MB | NPU
74
+ | NAFNet-DeBlur | ONNX | w8a16 | Snapdragon® X Elite | 123.321 ms | 148 - 148 MB | NPU
75
+ | NAFNet-DeBlur | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 83.717 ms | 4 - 1460 MB | NPU
76
+ | NAFNet-DeBlur | ONNX | w8a16 | Qualcomm® QCS6490 | 14323.605 ms | 1474 - 1492 MB | CPU
77
+ | NAFNet-DeBlur | ONNX | w8a16 | Qualcomm® QCS8550 (Proxy) | 118.494 ms | 0 - 109 MB | NPU
78
+ | NAFNet-DeBlur | ONNX | w8a16 | Qualcomm® QCM6690 | 7066.147 ms | 1459 - 1480 MB | CPU
79
+ | NAFNet-DeBlur | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 6463.166 ms | 1403 - 1420 MB | CPU
80
+ | NAFNet-DeBlur | ONNX | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 69.946 ms | 4 - 1052 MB | NPU
81
+ | NAFNet-DeBlur | ONNX | w8a16 | Qualcomm® QCS9075 | 145.941 ms | 4 - 49 MB | NPU
82
+ | NAFNet-DeBlur | ONNX | w8a16 | Qualcomm® QCS7790 | 6463.166 ms | 1403 - 1420 MB | CPU
83
+ | NAFNet-DeBlur | ONNX | w8a16 | Qualcomm® QCS8750 | 69.946 ms | 4 - 1052 MB | NPU
84
+ | NAFNet-DeBlur | ONNX | w8a16 | Qualcomm® QCS7181 | 123.321 ms | 148 - 148 MB | NPU
85
+ | NAFNet-DeBlur | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 44.478 ms | 0 - 991 MB | NPU
86
+ | NAFNet-DeBlur | QNN_DLC | float | Snapdragon® X2 Elite | 47.47 ms | 3 - 3 MB | NPU
87
+ | NAFNet-DeBlur | QNN_DLC | float | Snapdragon® X Elite | 112.202 ms | 3 - 3 MB | NPU
88
+ | NAFNet-DeBlur | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 80.681 ms | 3 - 1203 MB | NPU
89
+ | NAFNet-DeBlur | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 109.364 ms | 3 - 5 MB | NPU
90
+ | NAFNet-DeBlur | QNN_DLC | float | Qualcomm® SA8775P | 139.945 ms | 0 - 915 MB | NPU
91
+ | NAFNet-DeBlur | QNN_DLC | float | Qualcomm® SA8650P | 139.945 ms | 0 - 915 MB | NPU
92
+ | NAFNet-DeBlur | QNN_DLC | float | Qualcomm® SA8255P | 139.945 ms | 0 - 915 MB | NPU
93
+ | NAFNet-DeBlur | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 186.589 ms | 2 - 841 MB | NPU
94
+ | NAFNet-DeBlur | QNN_DLC | float | Qualcomm® SA8295P | 162.508 ms | 0 - 632 MB | NPU
95
+ | NAFNet-DeBlur | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 63.577 ms | 0 - 944 MB | NPU
96
+ | NAFNet-DeBlur | QNN_DLC | float | Qualcomm® QCS9075 | 152.183 ms | 5 - 11 MB | NPU
97
+ | NAFNet-DeBlur | QNN_DLC | float | Qualcomm® QCS8750 | 63.577 ms | 0 - 944 MB | NPU
98
+ | NAFNet-DeBlur | QNN_DLC | float | Qualcomm® QCS7181 | 112.202 ms | 3 - 3 MB | NPU
99
+ | NAFNet-DeBlur | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 40.146 ms | 1 - 1034 MB | NPU
100
+ | NAFNet-DeBlur | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 43.299 ms | 1 - 1 MB | NPU
101
+ | NAFNet-DeBlur | QNN_DLC | w8a16 | Snapdragon® X Elite | 91.135 ms | 1 - 1 MB | NPU
102
+ | NAFNet-DeBlur | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 62.882 ms | 0 - 1030 MB | NPU
103
+ | NAFNet-DeBlur | QNN_DLC | w8a16 | Qualcomm® QCS8275 | 192.965 ms | 2 - 781 MB | NPU
104
+ | NAFNet-DeBlur | QNN_DLC | w8a16 | Qualcomm® QCS8550 (Proxy) | 88.845 ms | 2 - 5 MB | NPU
105
+ | NAFNet-DeBlur | QNN_DLC | w8a16 | Qualcomm® SA8775P | 86.059 ms | 2 - 780 MB | NPU
106
+ | NAFNet-DeBlur | QNN_DLC | w8a16 | Qualcomm® SA8650P | 86.059 ms | 2 - 780 MB | NPU
107
+ | NAFNet-DeBlur | QNN_DLC | w8a16 | Qualcomm® SA8255P | 86.059 ms | 2 - 780 MB | NPU
108
+ | NAFNet-DeBlur | QNN_DLC | w8a16 | Qualcomm® QCM6690 | 1292.196 ms | 1 - 1488 MB | NPU
109
+ | NAFNet-DeBlur | QNN_DLC | w8a16 | Qualcomm® SA7255P | 192.965 ms | 2 - 781 MB | NPU
110
+ | NAFNet-DeBlur | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 137.101 ms | 1 - 1183 MB | NPU
111
+ | NAFNet-DeBlur | QNN_DLC | w8a16 | Snapdragon® 8 Elite For Galaxy Mobile | 52.163 ms | 1 - 934 MB | NPU
112
+ | NAFNet-DeBlur | QNN_DLC | w8a16 | Qualcomm® QCS9075 | 101.792 ms | 1 - 6 MB | NPU
113
+ | NAFNet-DeBlur | QNN_DLC | w8a16 | Qualcomm® QCS7790 | 137.101 ms | 1 - 1183 MB | NPU
114
+ | NAFNet-DeBlur | QNN_DLC | w8a16 | Qualcomm® QCS8750 | 52.163 ms | 1 - 934 MB | NPU
115
+ | NAFNet-DeBlur | QNN_DLC | w8a16 | Qualcomm® QCS7181 | 91.135 ms | 1 - 1 MB | NPU
116
+ | NAFNet-DeBlur | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 46.057 ms | 2 - 1076 MB | NPU
117
+ | NAFNet-DeBlur | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 80.313 ms | 0 - 1282 MB | NPU
118
+ | NAFNet-DeBlur | TFLITE | float | Qualcomm® QCS8275 | 483.489 ms | 3 - 1002 MB | NPU
119
+ | NAFNet-DeBlur | TFLITE | float | Qualcomm® QCS8550 (Proxy) | 109.824 ms | 3 - 7 MB | NPU
120
+ | NAFNet-DeBlur | TFLITE | float | Qualcomm® SA8775P | 139.931 ms | 3 - 1005 MB | NPU
121
+ | NAFNet-DeBlur | TFLITE | float | Qualcomm® SA8650P | 139.931 ms | 3 - 1005 MB | NPU
122
+ | NAFNet-DeBlur | TFLITE | float | Qualcomm® SA8255P | 139.931 ms | 3 - 1005 MB | NPU
123
+ | NAFNet-DeBlur | TFLITE | float | Qualcomm® QCS8450 (Proxy) | 206.205 ms | 0 - 1003 MB | NPU
124
+ | NAFNet-DeBlur | TFLITE | float | Qualcomm® SA8295P | 162.481 ms | 3 - 738 MB | NPU
125
+ | NAFNet-DeBlur | TFLITE | float | Snapdragon® 8 Elite For Galaxy Mobile | 63.17 ms | 2 - 1033 MB | NPU
126
+ | NAFNet-DeBlur | TFLITE | float | Qualcomm® QCS9075 | 153.064 ms | 1 - 150 MB | NPU
127
+ | NAFNet-DeBlur | TFLITE | float | Qualcomm® SA7255P | 483.489 ms | 3 - 1002 MB | NPU
128
+ | NAFNet-DeBlur | TFLITE | float | Qualcomm® QCS8750 | 63.17 ms | 2 - 1033 MB | NPU
129
+
130
+ ## License
131
+ * The license for the original implementation of NAFNet-DeBlur can be found
132
+ [here](https://github.com/megvii-research/NAFNet/blob/main/LICENSE).
133
+
134
+ ## References
135
+ * [Simple Baselines for Image Restoration](https://arxiv.org/abs/2204.04676)
136
+ * [Source Model Implementation](https://github.com/megvii-research/NAFNet.git)
137
+
138
+ ## Community
139
+ * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
140
+ * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).
release_assets.json ADDED
@@ -0,0 +1,46 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "version": "0.55.0",
3
+ "precisions": {
4
+ "w8a16": {
5
+ "universal_assets": {
6
+ "qnn_dlc": {
7
+ "tool_versions": {
8
+ "qairt": "2.45.0.260326154327"
9
+ },
10
+ "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_deblur/releases/v0.55.0/nafnet_deblur-qnn_dlc-w8a16.zip"
11
+ },
12
+ "onnx": {
13
+ "tool_versions": {
14
+ "qairt": "2.42.0.251225135753_193295",
15
+ "onnx_runtime": "1.25.0"
16
+ },
17
+ "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_deblur/releases/v0.55.0/nafnet_deblur-onnx-w8a16.zip"
18
+ }
19
+ }
20
+ },
21
+ "float": {
22
+ "universal_assets": {
23
+ "tflite": {
24
+ "tool_versions": {
25
+ "qairt": "2.45.0.260326154327",
26
+ "litert": "1.4.3"
27
+ },
28
+ "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_deblur/releases/v0.55.0/nafnet_deblur-tflite-float.zip"
29
+ },
30
+ "qnn_dlc": {
31
+ "tool_versions": {
32
+ "qairt": "2.45.0.260326154327"
33
+ },
34
+ "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_deblur/releases/v0.55.0/nafnet_deblur-qnn_dlc-float.zip"
35
+ },
36
+ "onnx": {
37
+ "tool_versions": {
38
+ "qairt": "2.42.0.251225135753_193295",
39
+ "onnx_runtime": "1.25.0"
40
+ },
41
+ "download_url": "https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/nafnet_deblur/releases/v0.55.0/nafnet_deblur-onnx-float.zip"
42
+ }
43
+ }
44
+ }
45
+ }
46
+ }